›_ Codex Issues
Codex Issues / #17327

Codex Desktop Stop/checkpoint can lose original thread closeout bubble after continuation

Open 💬 2 comments Opened Apr 10, 2026 by GongYuanCaiJi
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

Upstream Bug Report - Codex Desktop Stop/Checkpoint UI Loss

Summary

Pressing Stop can preserve the continuation checkpoint but remove the last visible assistant closeout bubble from the original thread.

Steps to Reproduce

  1. Start a long-running Codex Desktop task in a thread.
  2. While the assistant is still responding or wrapping up, press Stop.
  3. Let Codex create the continuation/checkpoint thread.
  4. Reopen the original thread.
  5. The continuation thread proves the Stop event happened, but the original thread no longer shows the last visible assistant closeout bubble.

Environment

  • macOS
  • Codex Desktop 26.406.31014 (1395)
  • app bundle path: /Applications/Codex.app

Expected

  • The original thread keeps the last visible assistant closeout bubble and the turn closes cleanly.

Actual

  • The continuation thread contains an auto-checkpoint for the same session, but reopening the original thread no longer shows the assistant closeout bubble the operator saw before switching away.

Additional Context

  • This report is based on one historical source thread with operator-corroborated visible content loss after Stop/continuation.
  • A broader local stop-capture dataset shows that raw open-turn-after-stop events are common in proof/probe interruptions and should not be read as bug frequency by themselves.
  • The continuation checkpoint survived; the missing piece is the original thread's visible closeout bubble.

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 4 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #16599
  • #16644
  • #15849

Powered by Codex Action

GongYuanCaiJi · 4 months ago

Checked the three duplicates flagged by the bot.

I think this issue is related to the same persistence/history family, but it is still distinct enough to keep open:

  • #16599 is broader restart-triggered history rollback / subagent loss after app restart.
  • #16644 is a closed RCA / root-cause-analysis issue about restart-time persistence flush, not a user-facing stop/continuation report.
  • #15849 is a stuck-"thinking" thread after restart, not a missing closeout bubble after Stop creates a continuation checkpoint.

This issue is narrower:

  • it is about a Stop/continuation flow
  • the continuation checkpoint survives
  • the missing piece is the original thread's last visible assistant closeout bubble
  • it does not require the whole thread or subagent history to disappear

If maintainers prefer to consolidate this into an existing user-facing history-loss issue, I am happy to move the evidence there.